Exercise Class 9
This exercise is based upon Chapter 7 of Delwiche and Slaughters The Little SAS book. We will study two dierent macros here. Take your time and be sure you understand every step; be sure to look at intermediate data sheets in the WORK library so that you understand the macros actions. The rst macro was introduced in class. When running this macro in Leconte 124, I needed to include .. in between &strata and csv in the rst INFILE statement; on my laptop, this wasnt necessaryI only needed one period. Ive included the version with two periods. You should be able to tell from the LOG le if the correct data set name (upurban.csv) was assigned by the macro; if not, adjust accordingly. The macro computes the probability that sampling units are included in a sample of size 2, when sampling without replacement with unequal sampling probabilities. In class, we had shown that if unit i has sampling weight i , then the probability it is selected would be:
P (i) = P (i|i selected on rst draw) + P (i|i selected on second draw) = i j i + j = i 1 + 1 j 1 j j=i j=i The strata macro includes a DATA step with a nested DO loop. It looks complicated, but the actual macro features are really not that sophisticated&n is used mostly to modify the DO loop upper limit, and &strata is used to modify le names, and thats about it! Download and save upurban.csvits referred to as the trac intensity data on the web page. You can also go ahead and run the entire macro; several datasets will be created. Be sure to step through the code and take a look at them: Data set c drops the county name. Data set d transposes the data so that trac intensities are more easily converted to propotions (remember that SAS handles row operations easier than column operations). Data set e creates proportions from the trac intensities. Data set one cleans up some variables. Data set two computes our sampling probabilities. Data set three cleans up after our calculations. Data set four recaptures the county names.
1
Notice that were using DO and not %DO here; thats because were looping within a DATA step, and hence dont need the added exibility. Just because were writing a macro doesnt mean we need to use macro statements indiscriminately. The rest of the macro is relatively routine. Data four looks a little oddwe use firstobs here because when this program was rst written, the Stat Lab used to only compute sampling probabilities for counties that had actually been sampled, and so we wouldnt necessarily read the entire list of county names. We dont use a match merge in final add to the county names back into the data set since we never changed the order of the variables. Add up the sampling probabilities for the three countieswhat do you think is the signicance of the sum? In our second macro, we want to perform a series of simple linear regressions. Our macro should be able to accept a single dependent variable and a list of independent variables, and then construct regressions on each independent variable in turn. Lets start with a straightforward example, and then add a couple important features on our way to our nished macro. Our data set consists of assessment tests for a group of students. Our dependent (response) variable will be read. For our rst attempt, well simply use some macro variables. Go ahead and run all the commands through: ... proc reg data=hsb2; model read=&indvar; run; We actually just read in the macro regall, but I dont want to use it just yet. The inclusion of &SYSDAY and &SYSDATE is a little gratuitous here. Otherwise, you see a typical use of %LET commands to perform a multiple linear regression. The %scan function is new; %scan(string,i) that can save element i from string, where the elements are separated by spaces (as we did here) or by delimiters. Delimiters can be specied as a third argument to %scan. To conduct 4 simple linear regressions on each of those 4 variables, we could construct a macro that accepted a dependent variable, and a set of independent variables. As a rst attempt, we would likely nd it easiest to also include the number of independent variabl...